How to cycle count a two-person store without closing the floor

You have two people, one of whom is you, and the floor does not stop. The last full count was either a Sunday you lost or a thing you have meant to do since spring. Meanwhile the system says three and you can see two. Closing for a day to count everything would fix nothing — it would be wrong again by Friday.

The short answer

Count a slice every week instead of everything once a year. Pick the slice by value and by movement. Reconcile the same day, while you still remember what happened. Write down the cause, not just the correction — an adjustment with no reason attached is a number you will have to invent again next quarter.

Why the annual wall-to-wall count is the wrong tool

A full count has one virtue: the moment it finishes, every number is true. By week three it is a historical document and you have no idea which numbers went bad, or when. It is also the most expensive day you will run: you close, or you count around customers and do both badly, and you check a thousand lines to find variances on a few dozen. Worse, those variances are six months stale, so the cause is unrecoverable — nobody remembers a Tuesday in March. You end up adjusting to match the shelf, which is not a correction. It is a surrender written into the ledger.

A cycle count inverts all of that. It is short enough to do between customers, it touches what is most likely to be wrong, and the variance is days old, so you can still find the cause. A count that produces a corrected number is worth very little. A count that produces a corrected process is worth doing weekly.

Choosing the slice

Slice three ways at once — by value, by movement, by risk — and let frequency fall out of where a line sits. This is a starting schedule for a few hundred SKUs across one or two locations. Adjust it to your floor, not to this week.

| Slice | What is in it | How often | |---|---|---| | A — high value | Top items by dollars on hand. Draw the line where about half your inventory value sits | Every 2 weeks | | B — the middle | Steady sellers, moderate value | Monthly | | C — the long tail | Accessories, parts, consumables, slow movers | Quarterly | | Every serialized unit | Anything with a serial number or its own unit record | Monthly, all of them | | Anything that moved | A sale, receipt, transfer or return since its last count | Add to the next count | | Anything you couldn't find | A customer asked, the system said yes, you could not put your hands on it | Same day |

Two rows do the real work. Every serialized unit, monthly is non-negotiable anywhere you sell units with a plate on the back: one missing unit outweighs a year of accessory drift, and you cannot detect its absence by reading a quantity. Anything that moved is the cheapest accuracy you will ever buy — movement is what breaks numbers, so counting what moved finds errors at a far higher rate than counting at random. The whole thing is 20–30 minutes, twice a week, before opening.

What a count needs before you start

1. A cutoff time, decided before you start, not during.
2. A defined slice — this aisle, these 40 SKUs, every serialized unit at this location. Not "let's see how far we get."
3. No receiving during the count. None. See the next section.
4. Two people for anything serialized or over a value threshold you set.
5. A phone camera. Photograph the serial plate rather than transcribing it. Transcribed serials are where duplicate-unit errors come from.
6. Somewhere to write the cause, not just the number.

The cutoff rule, stated once and obeyed

Receiving stops at 4pm. Every movement after 4pm belongs to the next count.

That is the whole rule, and most bad counts are a violation of it wearing a shrink costume. A pallet lands at 4:40, someone puts two units on the floor because a customer is waiting, and at 5:15 your count finds two units the system does not know about. That is not shrink. It is a timing problem, and you will spend an hour proving it.

The rule cuts both ways, and the second half gets forgotten. A sale rung at 4:50 for a unit still on the floor because the customer collects Saturday is also a cutoff problem — the system has moved it and your eyes have not. Tag it and count it as sold. If a delivery cannot wait, park it in a marked corner, do not receive it, do not count it, deal with it tomorrow.

Counting blind, or counting against a sheet

Counting blind means the counter never sees the system quantity; the comparison happens afterwards. Blind is more accurate, and it is not close. Show someone a "3" and they will find three: they count the two in front, see a box behind, and write 3 without moving it. The number on the sheet is an answer, and people stop looking once they have one.

The compromise for two people:

• Serialized units and the A slice: blind, always. Being wrong is expensive here, and there are few enough lines that the extra minute costs nothing.
• Bulk C-slice lines: against a sheet is fine. Low value at risk.
• Any line with a variance gets recounted blind, by the other person, before anybody adjusts anything. A good share of first-pass variances vanish on recount, and each is an investigation you did not run.

Serialized units and bulk quantities are two different jobs

Do not run them in one pass. A bulk line is a quantity question. How many are here? One number, write it down, move on. A serialized unit is three questions, and quantity is not one of them:

1. Existence — is this specific unit here?
2. Location — is it where the system says, or in the back, in the van, at the shop, on a customer's hold?
3. Condition — is it still the grade it was received at? A unit shopworn, dented in handling or robbed for a part is not the unit the system thinks it has, even though it is right there.

Mixing the passes means doing the serial job at bulk speed, which means confirming units by the carton label instead of the plate on the machine. The two disagree more often than people think.

A variance is a question, not an answer

When the number is wrong, do not adjust. Investigate in this order, which is roughly how often each is the cause:

1. Recount it. Blind, by the other person, including the stack behind the stack and the shelf above eye level. Most variances die here.
2. Look for an unrecorded movement. A sale not rung. A receipt not posted. A transfer entered on one end only. A return taken over the counter and never keyed in. Check the line's movement history against the last few days of paper.
3. Look for the unit somewhere else. In the van from Thursday's delivery. On the bench in the shop. In the back on a customer hold. A unit in the wrong place reads exactly like a missing one and is a different problem.
4. Check whether it was scrapped or cannibalised. Somebody took the door off it in June and never wrote the movement.
5. Only then, theft. Last, deliberately. It happens, but it is the least common explanation for a single-unit variance and the most expensive one to assume. Jumping to it poisons the room and stops the search that would have found the unposted receipt.

Cause found: fix it, and post the correction with the cause attached. Cause not found after a real look: post it and record it as unexplained. Shrink is the unexplained difference — not every difference, just the residue after you rule the others out. Call every variance shrink and you learn nothing except that you are annoyed.

The adjustment, and when to stop looking

One entry. A quantity, a reason, a name, a date. An adjustment without a reason is worse than no adjustment, because it looks like information.

You also have to know when to stop. A workable rule: spend up to 15 minutes, or the value of the line, whichever is smaller. A $4 accessory is not worth a 40-minute search, however irritating — adjust it, mark it unexplained, move on. A $900 unit is worth as long as it takes, and if you cannot explain it today, leave the line open and come back tomorrow. Sometimes it turns up in the van.

What matters is not any single variance but the same SKU, shelf or section coming back wrong month after month. One unexplained unit is noise. The third on the same line is a process problem, and the process is the thing to fix.

What goes wrong

Receiving during the count. The most common way a count produces garbage. Symptom: the variances are all positive, all from one vendor, all found late.

One person counting their own section. Not dishonesty — familiarity. You already know what is on that shelf, so you verify your memory instead of the shelf. Swap sections.

Adjusting to match and never asking why. The count becomes a ritual that makes the system agree with the floor and teaches nothing. Symptom: every session fully adjusted, and not one note in any of them.

Counting the carton, not the plate. Symptom: duplicate serial numbers, or a unit that stays "present" until a customer wants it.

Letting the slice drift. Week one you count the A slice properly. Week six you count whatever is near the door. Hold the schedule, or admit you are doing an annual count in instalments.

How UpChannels handles this

Counts in UpChannels open as a session against one location, so a session holds the slice you counted, not the catalog. A search box narrows both lists by product name or SKU, and the trackable list by a unit's own serial number or barcode as well. Bulk lines and trackable units sit in separate lists, because they are separate jobs: a bulk line takes a quantity, a trackable unit is confirmed present or not present, one at a time. Units already Reserved, Sold, Used or Scrapped are locked there — their status comes from their order, not from a count.

Each line shows the variance against the system number and carries a note field. Use it. Nothing touches stock until you post the session, and posting writes one signed adjustment movement per line that has a variance, sets on hand to what you counted, recomputes available and closes the session — one transaction, once. Lines that matched are skipped.

Underneath, the movement ledger is append-only. Nobody using the app can delete a stock movement; a correction is a new compensating entry, so every quantity on screen traces back to the movements that produced it. A nightly job records the same integrity checks for every organization, so drift shows up as a dated row instead of a discovery six weeks later. And the Inventory page runs a one-click audit over your active stock levels for specific faults — ledger drift, negative on hand, more reserved than you own, unit records that disagree with the quantity. Where two readings are both possible you pick the true one, and the corrections you tick post as adjustment movements, not silent edits.

Two boundaries. The count screen shows the system quantity while you count, so a genuinely blind count means counting on paper first and keying it in afterwards. And the audit can tell you the ledger disagrees with the shelf; it cannot tell you why a unit is missing. That is still the list above.